From bulk descriptions to emergent interfaces: connecting the Ginzburg-Landau and elastic line models
Nirvana Caballero, Elisabeth Agoritsas, Vivien Lecomte, Thierry Giamarchi
FFrom bulk descriptions to emergent interfaces: connecting the Ginzburg–Landau andelastic line models
Nirvana Caballero , ∗ Elisabeth Agoritsas , Vivien Lecomte , and Thierry Giamarchi Department of Quantum Matter Physics, University of Geneva,24 Quai Ernest-Ansermet, CH-1211 Geneva, Switzerland Institute of Physics, Ecole Polytechnique Fdrale de Lausanne (EPFL), CH-1015 Lausanne, Switzerland and Universit Grenoble Alpes, CNRS, LIPhy, 38000 Grenoble, France (Dated: June 29, 2020)Controlling interfaces is highly relevant from a technological point of view. However, their richand complex behavior makes them very difficult to describe theoretically, and hence to predict.In this work, we establish a procedure to connect two levels of descriptions of interfaces: for abulk description, we consider a two-dimensional Ginzburg–Landau model evolving with a Langevinequation, and boundary conditions imposing the formation of a rectilinear domain wall. At thislevel of description no assumptions need to be done over the interface, but analytical calculationsare almost impossible to handle. On a different level of description, we consider a one-dimensionalelastic line model evolving according to the Edwards–Wilkinson equation, which only allows oneto study continuous and univalued interfaces, but which was up to now one of the most successfultools to treat interfaces analytically. To establish the connection between the bulk description andthe interface description, we propose a simple method that applies both to clean and disorderedsystems. We probe the connection by numerical simulations at both levels, and our simulations, inaddition to making contact with experiments, allow us to test and provide insight to develop newanalytical approaches to treat interfaces.
I. INTRODUCTION
Diverse systems including ferroic domain walls [1–11],cell fronts [12, 13], bacterial colonies [14], or contactlines [15] exhibit emergent structures separating different“states” or domains ( i.e. , different magnetization orien-tations in the case of ferromagnetic systems, or differentpolarization orientations in the case of ferroelectrics, orcells-media in cell fronts, or wet from dry in the caseof contact lines), usually called interfaces. From a tech-nological point of view, controlling interfaces is of greatinterest for various reasons. In some cases, interfaces areused as the base unit of devices (for example, in datastorage devices [16]), and in others, interfaces are usedto extract information about the whole system by sim-ply observing a fraction of the system (for example in thecase of cells colonies, where the interface gives informa-tion about the interactions present in the tissue [17]).Interfaces have been usually described as disorderedelastic systems (DES) [18, 19]. In this framework, in-terfaces are approximated by univalued and continuousfunctions of position and time. In a great number ofcases this is a good approximation since usually the re-gion where the system changes from a state to another issmall compared to the regions where the system is homo-geneous. In particular, in the aforementioned systems,interfaces can be treated as unidimensional elastic ob-jects, leading to a very simplistic description, which stillcaptures the essential ingredients describing the physicsof these objects. ∗ Corresponding author: [email protected]
The advantage of treating interfaces as one-dimensional univalued functions is that it allowsone to compute analytically, and in a very precise way,several observables and critical exponents describingdynamic and static properties of interfaces, allowing fora better understanding of their properties, and thus abetter control over them. However, it is well known thatreal experimental realizations of interfaces are usuallyfar away from being described by univalued functions,and in order to use the DES theoretical framework,uncontrolled approximations are used to force the realinterface to be adapted to one of the main hypotheses ofthis framework.On a different level of treatment for interfaces,Ginzburg–Landau (GL) models, where the state of thesystem is described by a local order parameter whichcan take real values in a well-defined range, can also de-scribe interfaces, and the advantage is that assumptionsabout the function describing the interface are no longerneeded. Moreover, effects like nucleation, bubbles, andnon-univalued interfaces may arise, allowing for a morerealistic description of interfaces. The lack of intrinsic pe-riodic pinning, usually present in spin-like models, makesthis approach extremely suitable for the study of inter-faces. However, analytical calculations are very difficultto tackle for these kind of models.Both levels of description, the elastic line model, andGL models have been proven helpful to describe thephysics of disordered systems very well. However, a com-plete connection between the two levels of description, or‘model reduction’, is still lacking. Establishing a connec-tion between both models is extremely important, sinceit allows one to obtain analytical predictions for the morecomplex model, based on results for its simpler counter- a r X i v : . [ c ond - m a t . d i s - nn ] J un part. This question is quite generic since the dynamicsis that of the so-called ‘model A’ [20]. A model reduc-tion has been determined for flat walls in the absence ofnoise [21], or using a Fokker–Planck viewpoint [22, 23]or other approaches for flat interfaces [24, 25], and inthe context of kinetic roughening [26] or of the ‘drum-head model’ [27, 28]. More complex approaches thanthe ones we propose have also been developed, includ-ing for instance the effect of curvature [25, 26, 29, 30] orof varying domain-wall width [27]. Note that the modelreduction is formally equivalent to the determination ofextended particle states in quantum field theory [31, 32],where collective coordinate methods are similar to thoseof statistical mechanics.In this work, we connect these two models througha simple procedure that requires few assumptions, andthat applies both to clean systems and to systems withquenched disorder. This is a first step to get insight inhow to extend the DES theory beyond the elastic ap-proximation, thus allowing for a better characterizationand understanding of experimental realizations of inter-faces. The plan of the paper is as follows. In Sec. II,we briefly describe the GL model, establish the neces-sary assumptions and propose a procedure to connectthis model to an Edwards–Wilkinson (EW) elastic linemodel, in the clean case. Complementary justificationsof our procedure are presented in Appendices A to C. InSec. III we compute analytically how the roughness, anobservable measuring geometrical fluctuations of an in-terface, evolves as a function of lengthscale and time fora 1D elastic line. We probe the established connectionbetween the models by performing simulations on a 2D-GL model, a 1D-EW model: we evaluate the roughnessof interfaces which evolved starting from a completelyflat configuration, and show how interfaces in both mod-els, under our proposed connection, behave in excellentagreement with the analytical prediction in the 1D case.We also probe the connection between models numeri-cally as a function of temperature. In Sec. IV, we intro-duce quenched disorder in the GL system and show howit translates in the EW model into a short-range corre-lated disorder. We evaluate numerically the roughnessand its Fourier transform, the structure factor, and showthat they are in excellent agreement in both models, val-idating our proposed procedure for disordered systems.We finally conclude and discuss some perspectives of ourwork in Sec. V. II. FROM BULK DYNAMICS TO INTERFACEDYNAMICS (CLEAN SYSTEMS)
We study the behavior of the region (or ‘interface’) sep-arating two domains characterized by distinct values ofthe local order parameter in a bulk model (see Fig. 1). Atthe bulk level, we use a Ginzburg–Landau (GL) modelto describe the system, where the order parameter ofeach homogeneous region is a local minimum of the cor-
FIG. 1. Snapshot of part of a system after solving numer-ically the Langevin equation (see text) for a 2D Ginzburg–Landau model (Eq. (3), with η = α = δ = γ = 1, T = 0 . t = 10 ) to obtain the evolution of the order parameter ϕ ( x, y ). The obtained interface for this system is also shownin black. One of the fitted soliton profiles ϕ ∗ ( x ) (for fixed y ) ishighlighted in dashed blue line. On the inset: the hyperbolicprofile ϕ ∗ ( x ) from Eq. (7), its derivative (which characterizesthe ‘density’ of the interface), and three typical states in thelocal double-well potential. responding “ ϕ ” potential. We consider a non-conservedorder parameter, ϕ ( r , t ), describing the local state of thesystem ruled by a GL Hamiltonian H GL [ ϕ ] = (cid:90) d r (cid:104) γ |∇ r ϕ | + V ( ϕ ) − hϕ (cid:105) , (1)where r ∈ R n , and the ϕ potential V ( ϕ ) = − α ϕ + δ ϕ (2)with α > δ >
0, models the existence of two preferredvalues for ϕ : the minima of this double-well potential at ± ϕ = ± (cid:112) α/δ represent the two preferential states ofthe system, and h is an external applied field.In this section, to establish the procedure, we focus ona clean system. The effect of disorder, which is crucialfor experimental realization of interfaces, will be studiedin details in Sec. IV.The simplest equation describing the time evolutionof the non-conserved order parameter ϕ ( r , t ) in contactwith a thermal bath at temperature T is given by theoverdamped Langevin equation η∂ t ϕ = − δ H GL [ ϕ ] δϕ + ξ = γ ∇ r ϕ − V (cid:48) ( ϕ ) + h + ξ , (3)where ξ = ξ ( r , t ) is a Gaussian white noise with zeromean and two-point correlator (cid:104) ξ ( r , t ) ξ ( r , t ) (cid:105) = 2 ηT δ n ( r − r ) δ ( t − t ) , (4) η is the microscopic friction and γ the amplitude of theelastic cost associated to deformations of ϕ .Interfaces are defined as the region where the orderparameter shifts from a preferred value to another. Weare interested in studying interfaces in a 2D system with r = ( x, y ) (see Fig. 1). To do so, if the x and y axesare chosen so that the interface has a univalued shapeat x = u ( y, t ), a natural ansatz to describe the field is ϕ ( x, y, t ) = ϕ ∗ ( x − u ( y, t )), where the function ϕ ∗ de-scribes the switch from a preferred value of the order pa-rameter to another. Such an ansatz can only be approxi-mate since, at non-zero temperature, the actual shape ofthe switching profile actually depends on the y coordinateand presents fluctuations of thermal origin (see Fig. 1).We expect it to become correct at low temperature if thefunction ϕ ∗ is well chosen. As shown in Appendix A,the thermal fluctuations of the order parameter ϕ ( x, y, t )in each of the ± ϕ phases are negligible compared totheir mean value if the temperature is much lower than T (cid:63) = αγ/δ . We thus expect our analysis to be valid inthe regime T (cid:28) T (cid:63) (see Ref. [33] for a treatment of ther-mal fluctuations in the bulk). In order to determine aneffective equation of evolution for the so-called displace-ment field u ( y, t ), we substitute the ansatz into the bulkLangevin Eq. (3): − ηϕ ∗(cid:48) ∂ t u = γ (cid:16) ϕ ∗(cid:48)(cid:48) + ϕ ∗(cid:48)(cid:48) ( ∂ y u ) − ϕ ∗(cid:48) ∂ y u (cid:17) − V (cid:48) ( ϕ ∗ ) + h + ξ. (5)Physically, we expect that at low temperature the op-timal ϕ ∗ is a solitonic profile that minimizes the energyof the system at zero field h : − δ H GL [ ϕ ] δϕ (cid:12)(cid:12)(cid:12) ϕ ∗ = γϕ ∗(cid:48)(cid:48) − V (cid:48) ( ϕ ∗ ) = 0 . (6)Such an equation effectively describes the conservativemotion of a “particle” of position ϕ ∗ and time x thatevolves in a potential V . If the function V ( ϕ ) has twolocal minima, we indeed have solitonic solutions that gofrom a minimum to another as x goes from −∞ to + ∞ .In our case of interest (2), we pick the soliton that sat-isfies the Dirichlet boundary conditions ϕ ∗ ( ±∞ ) = ∓ ϕ whose explicit form is well known: ϕ ∗ ( x ) = − ϕ tanh (cid:16) xw (cid:17) , (7)as illustrated in Fig. 1. The parameters w , represent-ing the width of the interface, and ϕ , representing thepreferred values ± ϕ for the order parameter are givenby ϕ = (cid:114) αδ , w = (cid:114) γα . (8) Substituting the identity (6) into Eq. (5), one obtainsexplicitly − ηϕ ∗(cid:48) ( x ) ∂ t u ( y, t )= γ (cid:104) ϕ ∗(cid:48)(cid:48) ( x ) (cid:0) ∂ y u ( y, t ) (cid:1) − ϕ ∗(cid:48) ( x ) ∂ y u ( y, t ) (cid:105) + h + ξ ( x + u ( y, t ) , y, t ) , (9)where we can safely replace ξ ( x + u ( y, t ) , y, t ) by ξ ( x, y, t )using the invariance by translation of the noise distribu-tion.The equation of evolution (9) is inconsistent (the de-pendency in x is not the same for every term), even atzero temperature. To obtain an equation of evolution forthe position of the interface, one multiplies Eq. (9) by ϕ ∗(cid:48) ,in order to “localize” the equation around the position ofthe interface, and one integrates over x . A justification ofthis procedure is presented in Appendix B (see Eq. (B9)):at the energetic level, when computing the force as de-riving from a bulk or an effective Hamiltonian, a factor ϕ ∗(cid:48) naturally appears between the derivatives δδu or δδϕ ∗ u .See also Appendix C for a path-integral approach wherethe integration over x comes naturally, directly in a dy-namical formulation. Doing so, one obtains η N ∂ t u = γ N ∂ y u − γ N ( ∂ y u ) + h N + ˜ ξ ( y, t ) , (10)where N ≡ (cid:90) ∞−∞ d x ( ϕ ∗(cid:48) ) = ϕ w = 2 √ δ (cid:115) α γ , (11) N ≡ (cid:90) ∞−∞ d x ϕ ∗(cid:48)(cid:48) ϕ ∗(cid:48) = 0 , N = (cid:90) ∞−∞ d x ϕ ∗(cid:48) = − ϕ . (12)The effective noise˜ ξ ( y, t ) = (cid:90) ∞−∞ d x ξ ( x, y, t ) ϕ ∗(cid:48) ( x ) (13)is a linear superposition of Gaussian variables, and is thusalso a Gaussian white noise of zero average and correla-tions (cid:104) ˜ ξ ( y , t ) ˜ ξ ( y , t ) (cid:105) = 2 ηT N δ ( y − y ) δ ( t − t ) . (14)We thus find a Langevin equation for u ( y, t ) of the form˜ η∂ t u = c∂ y u + F + ˜ ξ, (15)which is the EW equation [34] describing the time evolu-tion of an elastic line u ( y, t ), with friction ˜ η , elasticity c ,external force F , and temperature T . By this procedure,we found the friction and the force effectively “felt” by aninterface in the GL model, as well as its elastic constant,and how these quantities are related with the model pa-rameters as ˜ η ≡ η N = η √ αδ (cid:114) αγ ,c ≡ γ N = 2 √ αδ √ αγ ,F ≡ h N = − (cid:114) αδ h. (16)Note that the sign of the drive F does depend on the ex-plicit choice of soliton in Eq. (7): this is expected becausethe GL field h favors the + ϕ phase and will act with op-posite sign on the other possible soliton + ϕ tanh( x/w ).On the other hand, ˜ η and c are always defined as posi-tive, and their numerical prefactors depend on the spe-cific normalized density of the interface ρ w ( x ) ∝ | ϕ ∗(cid:48) ( x ) | (see Appendix D).By using the solitonic profile ϕ ∗ (Eq. (7)) as an ansatzto solve the Langevin equation for the GL model, wefound a procedure to go from the two-dimensional de-scription of the problem to an effective one-dimensionalone. Interestingly the same relation between the elastic-ity c of a domain wall in a one-dimensional system andthe GL parameters can be obtained by computing theenergy cost E el of the creation of a domain wall in thesystem, as was obtained before (see e.g. [21]).In this section we showed how to connect the GLand the DES descriptions at the level of their respec-tive Langevin equation. Our approach complements theone proposed in Ref. [35] where both the elasticity andthe thermal noise are also taken into account, but witha much more phenomenological treatment of the effec-tive thermal noise. The method we propose provides uswith an effective reduced dynamics for the interface dis-placement field u ( y, t ), that we test numerically in thesubsequent sections, on the evolution of roughness start-ing from a flat initial condition. We will discuss thisprocedure in presence of disorder in Sec. IV.In Appendix B we present a generic discussion on themodel reduction from an equilibrium Hamiltonian view-point that complements the dynamical approach pre-sented in this section. We show that the connectionbetween the GL and the DES descriptions can actuallybe performed directly at the level of the Hamiltonian aswell, if the system is assumed to be at equilibrium. Thisis thus relevant for the long-time limit of the equilibriumdynamics ( i.e. Eq. (3) with no external field h = 0), forwhich the probability of a given profile ϕ is simply givenby a Gibbs–Boltzmann distribution. This procedure onthe statics allows us to identify the DES elastic constant c and the effective disorder, but it does not give us access tothe effective DES friction and noise since those pertain tothe dynamics , so we need to consider the Langevin equa-tion as we did in this section (see also Appendix C). Notealso that the passage from Eq. (9) to Eq. (10) bears sim-ilarity with the projection operator of Refs. [25–27, 36]. III. ROUGHNESS OF INTERFACES
Among the observables that characterize interfaces,one of the most useful, convenient, and studied is theone that measures the spatial correlations of the position u ( y, t ) of the interface at time t , B ( r = | y − y | , t ) = (cid:104) [ u ( y , t ) − u ( y , t )] (cid:105) . (17)This so-called roughness function characterizes the ran-dom geometry of the interface. (cid:104) · · · (cid:105) denotes thermal av-erage, and · · · denotes the average over different disorderrealizations when appropriate. Usually, it is also conve-nient to compute the Fourier transform, called structurefactor, defined as S ( q, t ) = (cid:68) L u ∗ q ( t ) u q ( t ) (cid:69) , (18)where u q ( t ) = (cid:80) L − j =0 ( u j ( t ) − ¯ u ( t )) e iqj (¯ u ( t ) is the meanposition of the whole interface, zero thereafter), and thediscrete Fourier modes q = 2 πn/L with n = 1 , . . . , L − t goes toinfinity, if the interface has a finite length, correlationsspread along the whole interface, and this memory of theinitial condition disappears.For the clean system we are considering so far, we cancompute analytically the full time dependence of this cor-relation. One uses the linearity of the EW equation tosolve Eq. (15) for F = 0 [34], with an initially flat config-uration. Averaging over the thermal noise, one obtains: B ( r, t ) = T rc (cid:20) − √ πzr (cid:16) e − z r − (cid:17) − √ π (cid:90) zr e − t dt (cid:21) (19)where z = (cid:113) ˜ η ct . At large times, Eq. (19) converges tothe static thermal roughness B th ( r ) ≡ T r/c .We now use the result of Eq. (19) to assess the va-lidity of our bulk-to-line model reduction. To comparethe numerical efficiency of the 2D-GL and of the 1D-EW modelisations, we first perform simulations of the1D interface, i.e. we solve numerically Eq. (15) [37] withparameters ˜ η = c = √ (taking η = α = δ = γ = 1in Eq. (16)), T = 0 .
05, and F = 0 [38]. Starting from aflat configuration, we perform simulations of the elasticline during different times for different realizations. Foreach final configuration obtained for u ( y, t ) we compute B ( r, t ). In Fig. 2 we show the obtained roughness func-tions for each realization and for an average of B ( r, t ) overdifferent realizations. We find an excellent agreement be-tween the numerically obtained roughness functions andthe analytical result (19).The analytical prediction for the roughness functiongiven by Eq. (19) gives us a benchmark to test the FIG. 2. Time dependence of the roughness B ( r, t ), computed for interfaces in a 2D Ginzburg–Landau system (bottom figures)and for an equivalent 1D Edwards–Wilkinson system (top figures), obtained for 10 realizations (left figures) and for the averageover 10 realizations of simulations which evolved during a time t = 10 j , j = 1 , · · · , B ( r, t ) (Eq. (19)) for an equivalent one-dimensionalinterface is shown on dashed colored lines for different evolution times. The asymptotic value Tc r , expected for a completelystationarized interface, is shown in black dotted lines. On the right, the final extracted interfaces for one of the realizationsafter each evolution time t are shown for both models. A portion of length 25 . × . t = 10 and t = 10 , along with the detected interface.FIG. 3. Temperature dependence of the roughness B ( r, T ) for a 2D Ginzburg–Landau system (bottom figures) and for anequivalent 1D Edwards–Wilkinson system (top figures), obtained for 10 realizations (left figures) and for the average over 10realizations of simulations which evolved during a time t = 10 at temperatures T = 0 . , . , . , . , . , . B ( r, T )(Eq. (19)) for an equivalent one-dimensional interface is shown on dashed colored lines for different temperatures. The finalinterfaces obtained for one realization are also shown for both models at different temperatures. A portion of the Ginzburg–Landau system is also shown at T = 0 .
05 and T = 0 .
3, along with the detected interface, shown in black. proposed connection between the GL model of Eq. (3)and the EW dynamics of Eq. (15) in two and one di-mensions respectively. We performed simulations of a2D-GL system, by solving numerically Eq. (3), with α = δ = γ = η = 1, at T = 0 .
05, with periodic bound-ary conditions along y (interface direction), and Dirichletboundary conditions along x (see Fig. 1) [39].Let us define for convenience the bulk order parameter ϕ u ( x, y ) = ϕ ∗ ( x − u ( y )) associated to an interface of posi-tion u ( y ) and a solitonic profile given by Eq. (7) at each y .In the simulation, we start with a flat domain wall, i.e. with an initial condition ϕ ( x, y, t = 0) = ϕ u ( x, y ),with u ( y ) = L x /
2, for all y . The order parameter ϕ ( x, y, t ) then evolves in time by keeping the shape ofa rectilinear domain wall profile, localized along an in-terface of position u ( y, t ) (see Fig. 1).To obtain the effective interface position u ( y, t ) for agiven configuration ϕ ( x, y, t ) of GL model, we fit ϕ ( x, y, t )at fixed y and t with a function ϕ u ( x, y ), with the fittingparameters { ϕ , w, u ( y ) } . The interface position u ( y, t )is then given by the fitted value u ( y ) [40]. A snapshot ofpart of a simulated system is shown in Fig. 2 along withthe detected interface and some of the fitted interfacepositions u ( y, t ). By following this method, we computed u ( y, t ) for different realizations of simulations of a systemwhich evolved for different times, and we computed theroughness defined on Eq. (17) of these functions.The obtained values of the roughness are shown inFig. 2 for different realizations at each time, and alsofor the average of the roughness over different realiza-tions. The roughness functions of the interfaces obtainedin our simulations are in excellent agreement with theexpected result after different evolution times. For thepure system, this strongly supports that we have a veryprecise method to connect both levels of descriptions ofinterfaces, in the elastic approximation.This mapping allows us to test for the deviations forthe pure elastic description of the interface. For the 1D-EW model, where the elastic description is exact by con-struction, no deviation from the elastic description in-deed occurs. This can be seen in Fig. 3, where we com-puted the roughness of interfaces which evolved duringa time t = 10 for different temperatures T and com-pared it with the theoretical prediction (19) that we de-note B ( r, T ) to emphasize the temperature dependence.However, for the 2D-GL model, the measured roughnessfunctions match the predicted roughness only when theratio T /T (cid:63) is sufficiently small (see Appendix A), with T (cid:63) = αγ/δ = 1 for our parameter values. We observe de-viations from the theoretically expected value of B ( r, T )for temperatures larger than T = 0 .
15. Such discrepancyas temperature increases is expected, since the approachwe proposed to go from the bulk to the line model isbased on a small-noise hypothesis. u − u − . − . . . . . . F p ( u , y ) F p ( u , y ) Γ( u − u )average0 5 10 u − . − . . . . F p ( u , y ) FIG. 4. Computed correlations of 256 independent realiza-tions of pinning forces F p ( u, y ) (in gray). The average of thesecorrelations is plotted in pink, showing an excellent agreementwith the expected correlations given by Γ( u ) (Eq. (26)), shownin dashed black line. On the inset, 4 different realizations ofpinning forces are shown. IV. DISORDERED SYSTEMS
Disorder plays a key role inducing highly non-linear ef-fects in the statics and dynamics of interfaces. In partic-ular, it is well known that, as a consequence of disorder,the interface geometry is drastically changed comparedto one only subject to thermal fluctuations, and its studyis the whole point of the DES framework [9, 18, 19]. Atsmall lengthscales, thermal fluctuations are expected todominate the interface geometry behavior (at equilibrium B ( r ) ≈ B th ( r ) = Tc r ζ th , with ζ th = 1 / ζ are affected [19, 41].The equilibrium roughness B ( r ) will thus be character-ized at large distances by a different exponent depen-dent on the disorder type (for example random-bond orrandom-field types [42]). Let us now extend the mappingpresented on Sec. II to the case of disordered systems.To study the effect of quenched disorder on an inter-face described by a Ginzburg–Landau (GL) model, weintroduce fluctuations in the height of the double-wellpotential V ( ϕ ) of (2) as V ζ ( ϕ ( r )) = V ( ϕ ( r ))(1 + (cid:15)ζ ( r )) . (20)Here ζ ( r ) is a random number at position r taken from aGaussian distribution with zero mean and unit variance,whose correlations satisfy ζ ( r i ) ζ ( r j ) = δ ( r i − r j ), where r i,j are the relative distance between the simulation cells i and j , and we recall that · · · denotes the average overdifferent disorder realizations.When using the ansatz ϕ ( x, y, t ) = ϕ ∗ ( x − u ( y, t )), theLangevin equation describing the evolution of the orderparameter now becomes, instead of (9), − ηϕ ∗(cid:48) ∂ t u = γ (cid:16) ϕ ∗(cid:48)(cid:48) + ϕ ∗(cid:48)(cid:48) ( ∂ y u ) − ϕ ∗(cid:48) ∂ y u (cid:17) (21) − V (cid:48) ( ϕ ∗ ) − (cid:15)ζ ( x, y ) V (cid:48) ( ϕ ∗ ) + ξ ( x, y, t ) . Following the procedure of Sec. II, i.e. by multiplying by − ϕ ∗(cid:48) , using the soliton equation (6) γϕ ∗(cid:48)(cid:48) = V (cid:48) ( ϕ ∗ ), andintegrating x over the whole space, we find an effectiveLangevin equation for the displacement field u ( y, t )˜ η∂ t u = c∂ y u + F p ( u ( y, t ) , y ) + F + ˜ ξ ( y, t ) . (22)Compared to Eq. (15), we have now the extra term F p ( u, y ) = (cid:15)γ (cid:90) ∞−∞ d x ζ ( x + u, y ) ϕ ∗(cid:48)(cid:48) ( x ) ϕ ∗(cid:48) ( x ) , (23)which represents a quenched pinning force acting on theinterface. As a linear combination of a Gaussian field, therandom pinning force F p is again Gaussian. Its averageis zero and its correlations are given by F p ( u , y ) F p ( u , y ) = (cid:15) δ ( y − y )Γ( u − u ) , (24)where the correlator along the x direction is defined asΓ( u ) = γ (cid:90) ∞−∞ d x (cid:0) ϕ ∗(cid:48) ϕ ∗(cid:48)(cid:48) (cid:1) ( x ) (cid:0) ϕ ∗(cid:48) ϕ ∗(cid:48)(cid:48) (cid:1)(cid:0) x − u (cid:1) . (25)Using the explicit shape (7) of the profile ϕ ∗ ( x ), oneobtains by direct computationΓ( u ) = 2 α γ δ w sinh (cid:0) uw (cid:1) (cid:16)
115 sinh (cid:16) uw (cid:17) + 90 sinh (cid:18) uw (cid:19) + 7 sinh (cid:18) uw (cid:19) − uw
336 cosh (cid:16) uw (cid:17) − uw
81 cosh (cid:18) uw (cid:19) − uw (cid:18) uw (cid:19)(cid:19) . (26)The effective disorder correlations are thus short-rangewith a correlation length of the order of the interfacewidth w (see also Appendix D). The Fourier transform ofthe correlator (25), defined as ˆΓ( q ) = (cid:82) ∞−∞ d u e − iqu Γ( u ),is given by ˆΓ( q ) = Dg ( q, w ), where D = α γ δ and g ( q, w ) = π w ( wq ) (cid:0) w q + 4 (cid:1) sinh − (cid:16) πwq (cid:17) . (27) A pinning force with correlations given by Eq. (24),for fixed y and continuous u , may be generated by com-puting F p ( u, y ) = (cid:15) (cid:113) DL T (cid:80) M − n =0 e iq n u g ( q n , w ) z n , where q n = πL x n and z n are complex Hermitian random num-bers taken from a Gaussian distributions with zero meanand unit variance, with z =0. Here, L x = M δl is thetransverse length of the system. In Fig. 4 we show thecomputed correlations of pinning forces generated withthis method, for M = 10 , δl = 0 . D = 1, (cid:15) = 1 [43].In Fig. 5, we show the excellent agreement betweensimulations on the 2D-GL model and on the 1D-EWmodel where disorder was implemented through theaforementioned method. At large time and large scale,the roughness function departs from the thermal behav-ior ∼ r ζ th by developing a power-law regime which iscompatible with the expected scaling ∼ r ζ RB of the so-called ‘random-bond’ regime ( ζ RB = 2 / T /T (cid:63) ), allows one to avoid recom-puting dynamic and static exponents of interest for themore “realistic” GL case. More importantly, how differ-ent quantities deviate from the expected value when theelastic limit is not satisfied may be studied in detail.Besides, the mapping between the 2D-GL and the 1D-EW, when the elastic limit is satisfied, allows one to re-duce the system size from L x × L y to L y , and hencethe computational cost [44]. In addition, from a generalpoint of view, the model reduction allowed us to deter-mine explicitly the disorder distribution to which the GLinterface is effectively subjected to, as a result of thebulk disorder. We focused on the random bond case,but other cases, such as random-field or random-periodicdisorders [18] can be treated in a similar fashion as wedid. Generically, the method we propose allows one inprinciple to determine the effective disorder of the EWmodel starting from an arbitrary disorder distribution atthe bulk GL level. V. CONCLUSION AND PERSPECTIVES
Solving interface statics and dynamics beyond the elas-tic approximation is still a largely open theoretical andanalytical problem. The disordered elastic systems theo-retical framework has been proven helpful to analyze in-terface properties under the elastic approximation, but itcan not take into account many features of experimentalinterfaces. A more complete description, at a large com-putational cost, is to use directly the Ginzburg–Landau(GL) description of the whole system ( e.g. in 2D), where, r − B ( r , t ) ∼ r ζ RB Tc r ζ Th − − q − − S ( q , t ) Edwards–WilkinsonGinzburg–Landau Tc q − (1+2 ζ RB ) t FIG. 5. Comparison of observables for a 2D Ginzburg–Landau system (continuous lines) and for an equivalent 1DEdwards–Wilkinson system (dot-dashed lines), obtained af-ter averaging over 10 realizations of simulations which evolvedduring different times t , indicated by different colors, at tem-perature T = 0 .
05 and with disorder intensity (cid:15) = 0 . B ( r, t )for the larger simulation times show deviations from the ther-mal regime (dotted black line). For these larger times, B ( r, t )is characterized by the roughness exponent ζ RB = 2 /
3. Onthe bottom figure, we show the structure factor S ( q, t ), de-fined in Eq. (18). by opposition to the 1D elastic line model, no assump-tions need to be done over the function describing theposition of the interface.Connecting quantitatively these two descriptions hashowever proved elusive for extended interfaces, especiallyin presence of quenched disorder. We demonstrate in thepresent paper an analytical method to connect quanti-tatively the GL and the EW models with very simpleassumptions. Compared to historical approaches thatare either complex [22, 23, 26–30] or deal with rigidwalls [24, 25], or are more phenomenological [35], themethod we propose has the advantage of simplicity whileretaining the main features of the bulk dynamics. We testthis method by performing simulations at both levels in2D and 1D respectively, showing how an interface in theGL model behaves. We obtain an excellent agreementwith an effective elastic line in the EW model with theadequate elastic coefficients, friction and disorder distri-bution.In particular, we examine the evolution in time andspace of an evolving interface which is initially flat inboth models by computing its spatial correlations, theso-called roughness B ( r, t ), as a function of the evolvingtime of interfaces. For clean systems, we compute analyt-ically how the roughness B ( r, t ) of interfaces should be-have under the elastic approximation, and we show howthe simulated interfaces follow accurately our analyticalpredictions. We also probe the limit of the model reduc-tion (which is expected to be valid in the low-temperaturelimit) by showing that the dynamics of the GL interfacedeparts from the EW one at high enough temperature.We also determined the characteristic temperature T (cid:63) below which the effective 1D description is expected tobe valid.Our method, which has been demonstrated on thetime-dependent motion of a 1D interface, is quite generaland can be applied to other systems. The possibility to gofrom the GL to the much simpler interface has a twofoldinterest: (i) for systems for which the elastic limit is valid,it provides a path to speed up considerably the simula-tions compared to the higher dimensional GL descrip-tion, while retaining the semi-microscopic knowledge ofthe parameters of the system that are more readily ac-cessible from experiments for the GL description thanfor the more phenomenological interface one; (ii) for sys-tems for which the elastic limit is violated due to too largethermal noise or disorder strengths, it provides a path toquantitatively compare the direct GL simulation includ-ing all these effects with the simplified elastic description.This should help in asserting the role of “defects” such asoverhangs, bubbles or for periodic systems with topolog-ical defects. Our approach also gives a framework to testand develop new observables to study the geometry ofinterfaces with overhangs and bubbles. It also serves asa tool to test how the roughness of interfaces is affectedby defects.These exciting directions go clearly beyond the reachof the present paper and will be left for future studies. VI. ACKNOWLEDGMENTS
This work was supported in part by the Swiss NationalScience Foundation under Division II. N.C. acknowledgessupport from the Federal Commission for Scholarshipsfor Foreign Students for the Swiss Government Excel-lence Scholarship (ESKAS No. 2018.0636) for the aca-demic year 2018-19. V.L. thanks the Universit de Genve(where part of this work was performed) for its warmhospitality, and acknowledges support by the ERC Start-ing Grant No. 680275 MALIG, the ANR-18-CE30-0028-01 Grant LABS and the ANR-15-CE40-0020-03 GrantLSD. E.A. acknowledges support from the Swiss Na-tional Science Foundation by the SNSF Ambizione GrantPZ00P2 173962. We would also like to thank S. Bustin-gorry, J.-P. Eckmann, E.E. Ferrero, and A.B. Koltonfor fruitful discussions related to this work. We alsothank J.-P. Eckmann for a constructive criticism of themanuscript. The simulations were performed at the Uni-versit de Genve on the
Mafalda cluster.
Appendix A: Low temperature
In this Appendix, we determine the condition on thetemperature T which ensures that the thermal fluctu-ations of the bulk order parameter ϕ ( r , t ) around oneof the values ± ϕ remain small compared to the differ-ence of order parameter 2 ϕ between the two phases. Todo so, one can write ϕ ( r , t ) = (1 + ˆ ϕ ( r , t )) ϕ and de-termine in which regime of temperature ˆ ϕ ( r , t ) remainsmuch smaller than 1 far away from the domain wall po-sition. Expanding the Langevin equation (3) (in the ab-sence of external field h ), one finds ηϕ ∂ t ˆ ϕ = γϕ ∇ r ˆ ϕ − αϕ ˆ ϕ + (2 ηT ) ˆ ξ , (A1)where the rescaled white noise ˆ ξ ( r , t ) has correlations (cid:104) ˆ ξ ( r , t ) ˆ ξ ( r (cid:48) , t (cid:48) ) (cid:105) = δ n ( r (cid:48) − r ) δ ( t (cid:48) − t ). Going to Fourierspace for the spatial coordinates, we see that for eachmode q , the Fourier transform ˆ ϕ q verifies an Ornstein–Uhlenbeck [45] equation of the form ∂ t ˆ ϕ q = − η (cid:2) α + γ q (cid:3) ˆ ϕ q + (cid:104) Tηϕ (cid:105) ˆ ξ q , (A2)with (cid:104) ˆ ξ q ( t ) ˆ ξ q (cid:48) ( t (cid:48) ) (cid:105) = δ n ( q (cid:48) + q ) δ ( t (cid:48) − t ). Its equal-timecorrelation function at large times is known [45] and reads (cid:104) ˆ ϕ q ( t ) ˆ ϕ q (cid:48) ( t ) (cid:105) = T δ n ( q (cid:48) + q )(2 α + γ q ) ϕ for t → ∞ . (A3)(One finds the same result by using the Boltzmann weightand a Hamiltonian expanded quadratically close to ϕ ).Coming back to real space, for our case of interest n = 2, i.e. r = ( x, y ), we see that, in the steady state,the equal-time correlations are logarithmically divergent (with the distance) if evaluated at two closeby points: for t → ∞ and δ r →
0, one has (cid:104) ˆ ϕ ( r , t ) ˆ ϕ ( r + δ r , t ) (cid:105) = Tγϕ (cid:16) Constant + log δ r w (cid:17) . (A4)In order to still get a typical temperature scale, one cantake a vector δ r of norm of the order w = (cid:112) γ/α andone finds (cid:104) ˆ ϕ ( r , t ) ˆ ϕ ( r + δ r , t ) (cid:105) ∝ Tγϕ for t → ∞ , (A5)up to a numerical prefactor. Using the expression of ϕ ,we thus define a characteristic temperature T (cid:63) = αγδ (A6)such that for T (cid:28) T (cid:63) , the typical amplitude of the ther-mal fluctuations of ˆ ϕ are small. Note that, up to a numer-ical factor, one has T (cid:63) = w ∆ V with w = (cid:112) γ/α thelengthscale of elasticity (which also gives the domain-wallwidth) and ∆ V = V (0) − V ( ϕ ) = α / (4 δ ) the barrierof the ϕ potential. From the expression of the Hamilto-nian, we thus see that T (cid:63) is an energy, as expected.We also refer the reader to Ref. [33] for a study of theinfluence of bulk thermal fluctuations on the motion ofinterfaces. Appendix B: Solitonic ansatz in the Hamiltonian
Here we show how the connection between the GL andthe DES descriptions can actually be performed directlyat the level of the Hamiltonian as well, if the system isassumed to be at equilibrium.We recall that, for the boundary conditions that weconsider ϕ ( x ± ∞ , y ) = ∓ ϕ , the solitonic profile ϕ ∗ ( x )is the exact optimal profile at zero temperature, with-out disorder and in absence of external field ( T = 0, ζ ≡ h = 0). It satisfies the extremalization condition δ H GL [ ϕ, ζ ] /δϕ ( r ) | ϕ ∗ = 0, which translates for the Hamil-tonian (1) into the equation γ ∇ ϕ ∗ ( r ) = V (cid:48) ζ =0 ( ϕ ∗ ( r )).As we did in Sec. II. we consider from now onthe 2D solitonic ansatz ϕ u ( x, y ) = ϕ ∗ ( x − u ( y )), where γϕ ∗(cid:48)(cid:48) ( x ) = V (cid:48) ζ =0 ( ϕ ∗ ( x )) and ϕ ∗ ( x ± ∞ ) = ∓ ϕ , and ouraim is to compute explicitly the corresponding Hamil-tonian. Since our derivation is not specific to thedouble-well potential V ζ =0 ( ϕ ), we will keep ϕ ∗ ( x ) genericbut remembering whenever needed its explicit formfrom Eqs. (7)-(8), ϕ ∗ ( x − u ) = − ϕ tanh(( x − u ) /w )with ϕ = (cid:112) α/δ and w = (cid:112) γ/α . We will moreoverneed the following definitions of constants, slight gen-0eralisations of Eqs. (11)-(12): N ( u ) ≡ (cid:90) d x (cid:2) ϕ ∗(cid:48) ( x − u ) (cid:3) x ∈ R ) = N , N ( u ) ≡ (cid:90) d x ϕ ∗(cid:48) ( x − u ) ϕ ∗(cid:48)(cid:48) ( x − u )= (cid:90) d x ∂ x (cid:20) ϕ ∗(cid:48) ( x − u ) (cid:21) ( x ∈ R ) = N , N ( u ) ≡ (cid:90) d x ϕ ∗(cid:48) ( x − u ) ( x ∈ R ) = N , (B1)with N = 0 since ϕ ∗(cid:48) ( x → ±∞ ) = 0, and specifically forthe double-well potential N = ϕ /w and N = − ϕ (for the boundary conditions ϕ ∗ ( x → ±∞ ) = ∓ ϕ ). Weemphasize that we are able to get rid of the dependenceon u in Eq. (B1) if ϕ ∗(cid:48) ( x ) decays sufficiently fast withrespect to system size in the x -direction; this becomesexact for x ∈ R , but should be kept in mind otherwise.We compute explicitly the energy associated to theansatz ϕ u ( x, y ): H GL [ ϕ u , ζ ] = (cid:90) d y d y (cid:110) γ ∇ ϕ u ( x, y )] + V ζ ( ϕ u ( x, y )) (cid:111) = (cid:90) d y d x (cid:110) γ (cid:104) ( ∂ x ϕ u ( x, y )) + ( ∂ y ϕ u ( x, y )) (cid:105) + (1 + (cid:15)ζ ( x, y )) V ζ =0 ( ϕ u ( x, y )) (cid:111) = (cid:90) d y (cid:110) γ N ( u ( y )) (cid:104) ( ∂ y u ( y )) + 1 (cid:105) + (1 + (cid:15)ζ ( x, y )) V ζ =0 ( ϕ ∗ ( x − u ( y ))) (cid:111) ≡ (cid:90) d y (cid:104) c ∂ y u ( y )) + U p ( u ( y ) , y ) (cid:105) + C≡ H
DES [ u, U p ] + C . (B2)In the last two steps, we have identified the DES elasticconstant and the effective pinning potential, respectively: c ≡ γ N ,U p ( u, y ) ≡ (cid:15) (cid:90) d x ζ ( x, y ) V ζ =0 ( ϕ ∗ ( x − u )) , (B3)and an additive term independent of u thanks to x ∈ R : C ≡ (cid:90) d y (cid:20) γ N ( u ( y )) + (cid:90) d x V ζ =0 ( ϕ ∗ ( x − u ( y ))) (cid:21) = (cid:90) d y (cid:20) γ N + (cid:90) d x V ζ =0 ( ϕ ∗ ( x )) (cid:21) . (B4)Although for an infinite system size C might diverge, itis a well-defined finite constant for any finite system size,and as such it can be safely removed by normalization ofthe Gibbs–Boltzmann weight from the definition of theactual DES Hamiltonian H DES [ u, U p ]. Physically C cor-responds to the elastic energy associated to the gradient in the x direction ( ∝ ( ∂ x ϕ u ( x, y )) ) and the energy asso-ciated to the bare double-well potential V ζ =0 (since thetwo phases ± ϕ are of equal energy and the domain wallis spatially symmetric in x ); if we assume the same soli-tonic profile ∀ y , as we have done with the ansatz ϕ u ( x, y ),then these two contributions to the energy do not dependon u and thus are indeed irrelevant in an effective DESdescription of the system.The pinning potential U p ( y, u ) is linear in the underly-ing GL disorder ζ , consequently it inherits its Gaussiandistribution, with zero mean U p ( u, y ) = 0 and two-pointcorrelation: U p ( u, y ) U p ( u (cid:48) , y (cid:48) ) ≡ R w ( u, u (cid:48) ) δ ( y − y (cid:48) ) ,R w ( u, u (cid:48) ) ≡ (cid:15) (cid:20)(cid:90) d x V ζ =0 ( ϕ ∗ ( x − u )) V ζ =0 ( ϕ ∗ ( x − u (cid:48) )) (cid:21) = (cid:15) γ (cid:90) d x ϕ ∗(cid:48) ( x − u ) ϕ ∗(cid:48) ( x − u (cid:48) ) . (B5)We used in the last equality the defining relation γϕ ∗(cid:48)(cid:48) = V (cid:48) ζ =0 ( ϕ ∗ ) (but no need to specify V ζ ( ϕ )), and thisallows us to notice that R w ( u, u (cid:48) ) = R w ( u − u (cid:48) ). In or-der to reconnect with the pinning force F p ( u, y ) definedin Eq. (23), note that F p ( u, y ) = − ∂ u U p ( u, y )= (cid:15) (cid:90) d x ζ ( x, y ) V (cid:48) ζ =0 ( ϕ ∗ ( x − u ( y ))) ϕ ∗(cid:48) ( x − u )= (cid:15)γ (cid:90) d x ζ ( x, y ) ϕ ∗(cid:48)(cid:48) ( x − u ) ϕ ∗(cid:48) ( x − u ) ( x ∈ R ) = (cid:15)γ (cid:90) d x ζ ( x + u, y ) ϕ ∗(cid:48)(cid:48) ( x ) ϕ ∗(cid:48) ( x ) . (B6)And as for the force-force correlator (24), we have simi-larly: F p ( u, y ) F p ( u (cid:48) , y (cid:48) ) = ∂ u ∂ u (cid:48) U p ( u, y ) U p ( u (cid:48) , y (cid:48) )= − R (cid:48)(cid:48) w ( u − u (cid:48) ) δ ( y − y (cid:48) ) (24) ≡ (cid:15) Γ( u − u (cid:48) ) δ ( y − y (cid:48) ) , (B7)with the correlator Γ( u − u (cid:48) ) introduced and discussed inSec. IV.The bottom line of Eq. (B2) is that, with the soli-tonic ansatz ϕ u ( x, y ), the GL Hamiltonian reduces ex-actly (without any approximation) into a DES Hamilto-nian function of u ( y ), of elastic constant c and pinningpotential U p (with the two-point correlator R w ( x )): H GL [ ϕ u , ζ ] (cid:12)(cid:12)(cid:12) α,γ,δ,(cid:15) ≡ H DES [ u, U p ] (cid:12)(cid:12)(cid:12) c,R w ( x ) + C . (B8)This also implies that, if we need to determine the DESforce acting on the displacement field u ( y ) in its asso-ciated Langevin dynamics, we must use the functional1‘chain rule’ as follows: F p ( u ( y ) , y ) = − δ H GL [ ϕ u , ζ ] δu ( y )= (cid:90) d x (cid:90) d y (cid:48) (cid:20) − δ H GL [ ϕ u , ζ ] δϕ u ( x, y (cid:48) ) (cid:21) δϕ u ( x, y ) δu ( y (cid:48) ) δu ( y (cid:48) ) δu ( y ) (cid:124) (cid:123)(cid:122) (cid:125) δ ( y − y (cid:48) ) = (cid:90) d x (cid:20) − δ H GL [ ϕ u , ζ ] δϕ u ( x, y ) (cid:21) (cid:2) − ϕ ∗(cid:48) ( x − u ( y )) (cid:3) , (B9)which firmly supports our procedure to go from Eq. (9)to Eq. (15), namely to multiply by the profile density ϕ ∗(cid:48) ( x − u ( y )) and perform the integration (cid:82) d x ( . . . ).In addition, our physical motivation for even consid-ering H GL [ ϕ u , ζ ] is that, at sufficiently low temperature,the statistical average over thermal fluctuations shouldbe dominated by the optimal profile. In a nutshell, thisassumption can be formalized as follows ( O being an ob-servable without an explicit dependence on the disorder): (cid:104)O(cid:105) = (cid:90) D ϕ P [ ϕ, ζ ] O [ ϕ ] [ansatz ϕ u ( x,y )] ≈ (cid:90) D u P [ u, ϕ ∗ , ζ ] O [ u, ϕ ∗ ] (B10)with P [ ϕ, ζ ] ∝ exp (cid:8) − T H GL [ ϕ, ζ ] (cid:9) being the Gibbs–Boltzmann weight and thanks to our result (B8) P [ u, ϕ ∗ , ζ ] ∝ exp (cid:26) − T H GL [ u, ϕ ∗ , ζ ] (cid:27) (cid:12)(cid:12)(cid:12) α,γ,δ,(cid:15) ∝ exp (cid:26) − T H DES [ u, U p ] (cid:27) (cid:12)(cid:12)(cid:12) c,R w ( x ) . (B11)This model reduction of the equilibrium path integral, us-ing only the solitonic ansatz ϕ u ( x, y ), should be modifiedfor slightly higher temperature by taking into account atfirst the thermal fluctuations both in u around the ϕ u and of the profile ϕ ∗ itself.Finally, note that the derivation presented in this ap-pendix can straightforwardly be generalized to higher di-mensions r = ( x, y ) ∈ R d , for an interface parametrizedby a displacement field u ( y ) ∈ R along the direction ˆ x with y ∈ R d − the ‘internal’ coordinate in the plane ⊥ ˆ x .Using the solitonic ansatz ϕ u ( x, y ) = ϕ ∗ ( x − u ( y )), weobtain: H GL [ ϕ u , ζ ] = (cid:90) d y (cid:104) c ∇ y u ( y )) + U p ( u ( y ) , y ) (cid:105) + C≡ H
DES [ u, U p ] + C , − δ H GL [ ϕ u , ζ ] δu ( y ) = c ∇ y u ( y ) + F p ( u ( y ) , y )= c ∇ y u ( y ) − ∂ u U p ( u, y ) | u = u ( y ) , (B12)where the only modification in the two-point correlatorsfor u p and F p consists in replacing the 1D δ ( y − y (cid:48) ) byits multi-dimensional counterpart δ d − ( y − y (cid:48) ). Appendix C: Path-integral approach
For the Ginzburg–Landau Langevin dynamics (3) inabsence of disorder ( ζ ≡ , t f ] writes P [ ϕ ] ∝ e − S [ ϕ ] (C1) S [ ϕ ] = 14 ηT (cid:90) t f d t (cid:90) d x d y (cid:16) η∂ t ϕ + δ H GL [ ϕ ] δϕ (cid:17) , (C2)where the action S [ ϕ ] is given in its Onsager–Machlup form. Using the solitonic ansatz for ϕ ( x, y, t ), ϕ u ( x, y, t ) ≡ ϕ ∗ ( x − u ( y, t )), the action S [ ϕ u ] represents(through e − S [ ϕ u ] ) the weight of the profile ϕ u among ev-ery other possible profile ϕ ( x, y, t ). Integrating over thecoordinate x , one finds by direct computation that S [ ϕ u ] = 14 T η N (cid:90) t f d t (cid:90) d y (cid:26)(cid:18) η N ∂ t u − γ N ∂ y u − h N (cid:19) + 1645 V (cid:0) ∂ y u ) (cid:27) , (C3)where according to Eqs. (11)-(12) we have N = ϕ /w and γ N = V w , with V = α /δ the amplitude of the ϕ potential. The quartic term ∝ ( ∂ y u ) indicates thatsuch an action is not exactly in the expected form ofan action corresponding to a Langevin equation for theevolution of u ( y, t ) with a Gaussian white noise. A sim-ilar quartic term occurs when implementing such proce-dure for the noisy Landau–Lifschitz–Gilbert bulk dynam-ics [46]. Such supplementary terms remind us that thezero-noise ansatz profile ϕ u is not the exact profile of thebulk model: this corresponds to the fact, discussed inthe main text, that at the Langevin level, Eq. (9) is notexact. In Eq. (C3), for small displacements u , it can beneglected and the effective action for the position u ( y, t )of the interface reads S eff [ u ] = 14˜ ηT (cid:90) t f d t (cid:90) d y (cid:16) ˜ η∂ t u − c∂ y u − F (cid:17) . (C4)It corresponds to an Edwards–Wilkinson equation for u ( y, t ), of the form (15), with the effective friction co-efficient ˜ η , elasticity constant c and external force F asthe ones we found in Eq. (16) using the direct Langevinapproach. Appendix D: Interface normalized density
The solitonic ansatz ϕ u ( x, y ) = ϕ ∗ ( x − u ( y )) is at thecore of our procedure for connecting the GL to the DESdescription, with ϕ ∗ being the exact optimal profile atzero temperature, without disorder and in absence of ex-ternal field.2The derivative of this profile, ϕ ∗(cid:48) ( x ), can be interpretedstraightforwardly as an unnormalized ‘density’ of the do-main wall (or interface) between the two phases ± ϕ im-posed by the boundary conditions (as illustrated in theinset of Fig. 1). Its normalized counterpart is then de-fined as: ρ w ( x ) ≡ ϕ ∗(cid:48) ( x ) (cid:82) R d x (cid:48) ϕ ∗(cid:48) ( x (cid:48) ) (B1) = ϕ ∗(cid:48) ( x ) N = 1 w ρ ( x/w ) (7) = ϕ w cosh ( x/w ) (cid:124) (cid:123)(cid:122) (cid:125) = ϕ ∗(cid:48) ( x ) ϕ = 12 w cosh ( x/w ) , (D1)and the last line is specific to the double-well poten-tial V ζ =0 ( ϕ ). Thereafter we keep the profile ϕ ∗ andits corresponding density generic, but keeping in mindthat (i) ϕ ∗ is an odd function (inherited for instancefrom the symmetric double well), and (ii) it satisfies γ∂ x ϕ ∗ = V (cid:48) ζ =0 ( ϕ ∗ ). Note that we denoted in Ref. [47]this normalized density by ρ ξ ( x ) where ξ corresponds tothe effective ‘width’ and thus can be identified (up to anarbitrary numerical constant) with the parameter w .We can consequently rewrite, for {N , N } whose def-initions are recalled in Eq. (B1), and with |N | = 2 ϕ : N = N (cid:90) R d x [ ρ w ( x )] = N w (cid:90) R d˜ x [ ρ (˜ x )] N = N (cid:90) d x ∂ x (cid:20) ρ w ( x ) (cid:21) = N w (cid:90) R d˜ x ∂ ˜ x (cid:20) ρ (˜ x ) (cid:21) = 0 , (D2)so the specific functional form of ϕ ∗(cid:48) will fix the nu-merical factor (cid:82) R d˜ x [ ρ (˜ x )] but the overall dependence N ∝ N /w will not change.Using the definition of ρ w ( x ), the disorder correlator R x ( u, u (cid:48) ) can similarly be rewritten as R w ( u, u (cid:48) ) = (cid:15) γ N (cid:90) d x ρ w ( x − u ) ρ w ( x − u (cid:48) ) = (cid:15) γ N w (cid:90) d˜ x ρ (˜ x − u/w ) ρ (˜ x − u (cid:48) /w ) = 1 w R (cid:18) uw , u (cid:48) w (cid:19) , (D3)and we can further define the strength of disorder andthe normalized correlator, respectively, as: D norm ≡ (cid:90) R d x R w ( x ) ,R norm w ( u, u (cid:48) ) ≡ R w ( u, u (cid:48) ) /D . (D4)The strength of disorder becomes: D norm = (cid:15) γ N (cid:90) R d x d x (cid:48) ρ w ( x (cid:48) − x ) ρ w ( x (cid:48) ) = (cid:15) γ N w (cid:90) R d˜ x d˜ x (cid:48) ρ (˜ x (cid:48) − ˜ x ) ρ (˜ x (cid:48) ) (cid:124) (cid:123)(cid:122) (cid:125) =numerical prefactor (D5)which yields, for the normalized density associated to thedouble-well potential ρ (˜ x ) =
12 cosh (˜ x ) : D norm = (cid:15) γ N w
19 = 4 (cid:15) γ ϕ w = (cid:15) α γδ ,R norm1 (˜ u ) = 9 (cid:90) R d˜ x
12 cosh (˜ x − ˜ u ) 12 cosh (˜ x ) ,c ≡ γ N = 2 √ δ (cid:112) α γ . (D6)All this construction can nevertheless be generalized toother density functionals for ρ w ( x ), modifying only thedifferent numerical prefactors: although qualitatively ir-relevant, those are crucial for achieving a quantitativeagreement, as the one we have been seeking in this work. [1] S. Cherifi-Hertel, H. Bulou, R. Hertel, G. Taupier,K. D. H. Dorkenoo, C. Andreas, J. Guyonnet, I. Gapo-nenko, K. Gallo, and P. Paruch, Non-ising and chiral fer-roelectric domain walls revealed by nonlinear optical mi-croscopy, Nature Communications , 1 (2017).[2] B. Ziegler, K. Martens, T. Giamarchi, and P. Paruch,Domain wall roughness in stripe phase BiFeO thin films,Physical Review Letters , 247604 (2013).[3] P. Paruch and J. Guyonnet, Nanoscale studies of fer-roelectric domain walls as pinned elastic interfaces,Comptes Rendus Physique , 667 (2013).[4] J. Guyonnet, E. Agoritsas, S. Bustingorry, T. Gia-marchi, and P. Paruch, Multiscaling analysis of ferroelec-tric domain wall roughness, Physical Review Letters ,147601 (2012). [5] W. S. Torres, R. D. Pardo, S. Bustingorry, A. Kolton,A. Lemaitre, and V. Jeudy, Universal dimensionalcrossover of domain wall dynamics in ferromagnetic films,Physical Review B , 201201 (2019).[6] K.-s. Lee, C.-w. Lee, Y.-j. Cho, S. Seo, D.-H. Kim, andS.-B. Choe, Roughness exponent of domain interface inCoFe/Pt multilayer films, IEEE transactions on magnet-ics , 2548 (2009).[7] N. B. Caballero, I. F. Aguirre, L. J. Albornoz, A. B.Kolton, J. C. Rojas-S´anchez, S. Collin, J. M. George,R. D. Pardo, V. Jeudy, S. Bustingorry, et al. , Excessvelocity of magnetic domain walls close to the depinningfield, Physical Review B , 224422 (2017).[8] A. Magni, G. Durin, S. Zapperi, and J. P. Sethna, Visu-alization of avalanches in magnetic thin films: temporal processing, Journal of Statistical Mechanics: Theory andExperiment , P01020 (2009).[9] J. Ferr´e, P. J. Metaxas, A. Mougin, J.-P. Jamet, J. Gor-chon, and V. Jeudy, Universal magnetic domain wall dy-namics in the presence of weak disorder, Comptes RendusPhysique , 651 (2013).[10] M. P. Grassi, A. B. Kolton, V. Jeudy, A. Mougin,S. Bustingorry, and J. Curiale, Intermittent collective dy-namics of domain walls in the creep regime, Physical Re-view B , 224201 (2018).[11] D. Jord´an, L. J. Albornoz, J. Gorchon, C.-H. Lambert,S. Salahuddin, J. Bokor, J. Curiale, and S. Bustingorry,Statistically meaningful measure of domain-wall rough-ness in magnetic thin films, Physical Review B (2020).[12] M. Huergo, M. Pasquale, A. Bolz´an, A. Arvia, andP. Gonz´alez, Morphology and dynamic scaling analysisof cell colonies with linear growth fronts, Physical Re-view E , 031903 (2010).[13] N. E. Muzzio, M. A. Pasquale, P. H. Gonz´alez, and A. J.Arvia, Influence of individual cell motility on the 2D frontroughness dynamics of tumour cell colonies, Journal ofBiological Physics , 285 (2014).[14] J. A. Bonachela, C. D. Nadell, J. B. Xavier, and S. A.Levin, Universality in bacterial colonies, Journal of Sta-tistical Physics , 303 (2011).[15] S. Moulinet, A. Rosso, W. Krauth, and E. Rolley, Widthdistribution of contact lines on a disordered substrate,Physical Review E , 035103 (2004).[16] S. S. Parkin, M. Hayashi, and L. Thomas, Magneticdomain-wall racetrack memory, Science , 190 (2008).[17] Personal communication with G. Rapin, S. Brown, andP. Paruch.[18] T. Giamarchi, A. B. Kolton, and A. Rosso, Dynamics ofdisordered elastic systems, Lecture Notes in Physics ,91 (2006).[19] E. Agoritsas, V. Lecomte, and T. Giamarchi, Disorderedelastic systems and one-dimensional interfaces, PhysicaB , 1725 (2012).[20] P. C. Hohenberg and B. I. Halperin, Theory of dynamiccritical phenomena, Reviews of Modern Physics , 435(1977).[21] S. M. Allen and J. W. Cahn, A microscopic theory for an-tiphase boundary motion and its application to antiphasedomain coarsening, Acta Metallurgica , 1085 (1979).[22] K. Kawasaki, Dynamics of fluctuations in unstable sys-tems. I, Progress of Theoretical Physics , 410 (1977).[23] K. Kawasaki and T. Ohta, Kink dynamics in one-dimensional nonlinear systems, Physica A: StatisticalMechanics and its Applications , 573 (1982).[24] R. Bausch, V. Dohm, H. K. Janssen, and R. K. P. Zia,Critical dynamics of an interface in 1 + (cid:15) dimensions,Physical Review Letters , 1837 (1981).[25] M. Grant and J. D. Gunton, Temperature dependence ofthe dynamics of random interfaces, Physical Review B , 5496 (1983).[26] B. Grossmann, H. Guo, and M. Grant, Kinetic rough-ening of interfaces in driven systems, Physical Review A , 1727 (1991).[27] K. Kawasaki and T. Ohta, Kinetic drumhead model of in-terface. I, Progress of Theoretical Physics , 147 (1982).[28] H. W. Diehl, D. M. Kroll, and H. Wagner, The interfacein a Ginsburg-Landau-Wilson model: derivation of thedrumhead model in the low-temperature limit, Zeitschrift f¨ur Physik B Condensed Matter , 329 (1980).[29] R. Zia, Normal coordinates and curvature terms in an in-terface hamiltonian, Nuclear Physics B , 676 (1985).[30] K. R. Elder, M. Grant, N. Provatas, and J. M. Koster-litz, Sharp interface limits of phase-field models, PhysicalReview E , 021604 (2001).[31] J. L. Gervais, A. Jevicki, and B. Sakita, Perturbationexpansion around extended-particle states in quantumfield theory, Physical Review D , 1038 (1975).[32] C. G. Callan and D. J. Gross, Quantum perturbationtheory of solitons, Nuclear Physics B , 29 (1975).[33] R. Bausch, V. Dohm, H. K. Janssen, and R. K. P. Zia, Ef-fects of bulk diffusion on interfacial dynamics, Zeitschriftfr Physik B - Condensed Matter , 121 (1991).[34] S. F. Edwards and D. R. Wilkinson, The surface statisticsof a granular aggregate, Proceedings of the Royal SocietyA , 17 (1982).[35] S. A. Safran, P. S. Sahni, and G. S. Grest, Kinetics ofordering in two dimensions. I. Model systems, PhysicalReview B , 2693 (1983).[36] M. Grant and J. D. Gunton, Domain growth in therandom-field ising model: The breakdown of self-similarscaling in two dimensions, Physical Review B , 6266(1984).[37] E. E. Ferrero, S. Bustingorry, and A. B. Kolton, Non-steady relaxation and critical exponents at the depinningtransition, Physical Review E , 032122 (2013).[38] We consider a discrete mesh along the y -direction, withmesh-size (cid:96) = 1 for a system of length L = 4096, whilekeeping a continuum value of u . We use the Euler methodto integrate the equation with a time step ∆ t = 10 − .[39] For a system of length L x = 256, L y = 4096, with spatialmesh-size (cid:96) = 1 (in both directions). We integrate theequation by following a semi-implicit Euler method withintegration time-step ∆ t = 0 . { ϕ , w } exhibit at low tempera-tures a Gaussian distribution (with mean µ and standarddeviation σ ) around the equilibrium values (Eq. (8)). At T = 0 .
05 we find µ ϕ = 0 .
99 and σ ϕ = 0 .
01 for ϕ , and µ w = 1 . σ w = 0 .
25 for w . These means are in goodagreement with the expected values ϕ = 1 and w = √ ,042406 (2013).[42] P. Chauve, T. Giamarchi, and P. Le Doussal, Creep anddepinning in disordered media, Physical Review B ,6241 (2000).[43] In spite of the Fourier representation, the computationalcost of obtaining the quenched random force is very highsince a Fourier transform needs to be done for each sim-ulation cell at each simulation step (for every updatedvalue of u ). In order to reduce this computational cost,we generated 4096 independent pinning forces F p ( u, y )beforehand, where the u ’s are sampled on a grid with M = 10 and δl = 0 .
1. With these, we define an ap-proximate F p ( u, y ), updated for each value of u by linearinterpolation during the simulation. We checked numer-ically that the disorder generated in this way possessesthe expected correlator with satisfactory precision. For asystem of length L = 256 we checked that this methodproduces the same results within numerical errors com-pared to the full computation of the Fourier transform. [44] Compared to a 2D Ginzburg–Landau simulation (solv-ing numerically Eq. (3), with V ζ given by Eq. (20)), the1D Edwards–Wilkinson Eq. (22) is roughly 2 times fasterto solve for an equivalent evolution time t with the in-terpolation method for the disorder implementation. Weprovide this information only for practical and illustra-tive purposes, and we do not intend to discuss accelera-tion techniques in the present work. The 2 × speed-up ofcomputation is a lower bound, and a baseline for furtherimprovement of the codes.[45] G. E. Uhlenbeck and L. S. Ornstein, On the theory ofthe brownian motion, Physical Review , 823 (1930).[46] R. A. Duine, A. S. N´u˜nez, and A. H. MacDonald, Ther- mally assisted current-driven domain-wall motion, Phys-ical Review Letters , 056605 (2007).[47] E. Agoritsas, V. Lecomte, and T. Giamarchi,Temperature-induced crossovers in the static roughnessof a one-dimensional interface, Physical Review B ,184207 (2010).[48] E. A. Jagla, Numerical simulations of two-dimensionalmagnetic domain patterns, Physical Review E , 046204(2004).[49] N. B. Caballero, E. E. Ferrero, A. B. Kolton, J. Curi-ale, V. Jeudy, and S. Bustingorry, Magnetic domain wallcreep and depinning: a scalar field model approach, Phys-ical Review E97